Hallo Carsten,

first, the very best filtering methode would be a closed Faraday cage. Faraday cage should not be part of any internal electronic circuit. Means, do not use Faraday cage for ground return currents. All wires, all chips, yes, everything has to sit inside.
Only at one position 0V of circuit is allowed to have a connection to Faraday cage: At measuring cable entry point, where the cable goes into enclosure of electronic circuit. Connection from signal ground (0V of circuit) to enclosure has to be made to inner side of enclosure (Faraday cage). This is neccessary, because outer side of enclosure is hit by cellphone radiation, and so is contaminated with high frequency. If enclosure is thick enough, interference will
be dampened so highly (while penetrating enclosure wall), that at it's inner side interference is mostly diminished. So, everything is done, that PCB is connected to a 'cold' point of Faraday cage.
Any additional filtering has also to be referenced to this point. This is neccessary because at the range of frequency of interest any copper trace on PCB would add it's inductiviy to low-pass-filter and destroy high frequency dampening.
You will only have success when you follow these design rules. Frequency of interference of typical cellphone is far behond 1GHz, and choosing the wrong ground referencing for filtering will make any filtering success obsolete.
Sometimes even the use of a second metallic enclosure is neccessary, which has to be an PCB mounted one. You know, like TV-tuners look like. It's essential that this is completely soldered to PCB, directly on massplane of BOTH sides of PCB. Filtering then has to be connected at this inner metallic enclosure. Also, connection to outer enclosure (completely closed Faraday cage) has to be made here. Use as short as possible leads for this connection.
Probably you will not use any connector because of special connection needs of thermocouple: It's demanded that wires are connected to one thermalizing copper block and there's some 'ice-point-compensation' neccessary. But if you use any connector keep in mind this:
It's extremely important to use the right connector of measuring cable. It must allow shielding to be connected to enclosure via 360° contact. NEVER USE PIGTAIL CONNECTION HERE! Otherwise you will not have completely closed Faraday cage, and any kind of filtering becomes fruitless. If you must, buy an expensive connector.
By the way, standard XLR-connectors, widely used in professional music recording is NOT well suited. Industry just tries to make some improvements, but I don't know whether there is already some suitable on market.
Output signal should leave enclosure next to input cable, and not on other side of enclosure. Here you also need this expensive connector. Not because output signal is almost as sensitive to cellphone radiation as input signal, but just to achieve completely closed Faraday cage.
I know, it seems to be ridiculous to treat nearly DC signal of thermocouple like sensible TV-tuner signals. But it's not only the SIGNAL that defines how to mount and shield a circuit, but also the interference signal. And when this is high frequent then the same design rules are guilty.

Now, how to filter?
You need something which definitely is high ohmic at range of frequency of interest inside signal line and something which definitely is low ohmic bypassing interference to ground. But additional specifications must be fullfilled. Thermocouple wires must be connected to a thermalizing copper block.
My recommendation: Put copper block on PCB, just outside inner enclosure. Connect measuring wires to it. Directly behind this point, entry of inner enclosure should sit. Position filter directly behind copper block and directly in front of entry of inner enclosure, just outside inner enclosure. Cold-ice-compensation circuitry must also sit inside inner enclosure. If it's sitting outside here additional filtering is neccessary. May be it's a good idea to put copper block and compensation circuit in good thermal contact to wall of inner enclosure, while sitting copper block outside and compensation circuit inside. When wall is good grounded, both parts are well shielded from each other.
Filter parts must have all leads or contacts having at the same temperature as copper block, otherwise additional thermoelectric potentials are built. Surface mounted devices are highly recommended. For two reasons:

1. High frequency performance is much better.
2. When they are sitting on massplane (outside of inner enclosure!) they highly profit from thermalizing construction (enclosure walls soldered directly on massplane) and keeping same temperature as copper block is eased.

Actual filter components have to be optimized for best suppression. I would suggest LRC-low-pass-filter, but may be RC-filter is enough. C should be a parallel circuit of some identical capacitors. This reduces the risk of resonances and drastically decreases inductance. Less than 1nH is readily achievable. Try 2...3 10nF X7R, from each line to ground. Not between lines!
I do not know whether your compensation circuitry allows dc-resistance at measuring cables (AD594/AD595 would!). So if there is no problem use 10kOhm resistors. Also here 2...3 in series are optimum, because stray capacitance of this arrengement decreases.
When dc-resistance is not allowed, only ferrite beads suited for high DC-currents can be used. Try BL02RN2 from MURATA. They are excellent working even beyond 1GHz! But you can also stick some pure ferrite beads over end of cable lines. But keep in mind that high frequency performance can be quite different. There are some types on the market only being good below 100MHz, which are not well suited for this job, at all! There's only one disadvantage in using a LC-filter: There can be some resonance, where filter is not having any dampening, but where even some amplification is possible. For this reason some dc-resistance might be neccessary.
Resonace frequency can easily be calculated, assumed that inductivity of ferrite bead is specified. From datasheet for BL02RN2 approximately 2.2µH can be estimated.
So, Thomson formula gives:

f = 1 / 2 / pi / SQRT(L x C)

If resonance occur can also be estimated: Keep R bigger than:

R > SQRT(2 x L / C)

Then there is NO resonance!

Let's have some example:
Two BL02RN2 and three 10nF are used per line. Then resonance frequency is 440kHz and R should be greater than 17 Ohm.

Hope it helps...
Kai