First let us get some values. I am using 10kHz as a reference, with your desired precision all results, are the same for your other ranges.
10 kHz is the equivalent of a pulse of 100 uSec. 1 Hz (required precision) is 1/10.000th of that. 1/10.000th of 100 uSec is 10 nSec. 10nSec is the equivalent of 100 Mhz. So the above show that to achieve the precision you want you need the counter to roll at 100 MHz.
Since the '51 runs the timer/counter at 1/12 the clock, you would need a '51 clocked at 1.2 GHz and I doubt that one will be available anytime soon. Ok, so you need to run the counter outside the '51 at 100 MHz. Your only chance is a CPLD where you can keep all signals inside a chip, PCB traces behave VERY nastily at such speeds.
Now, the next thought would be to have the frequency to be measured interrupt the '51 that then would stop the external counter and read it. Oops, ther is a little problem here, interrupt latency variance. The interrupt latency variuance for a '51 is 0-48 clock cycles, which would give you a variance of 0-4 uSec with a 12MHz xtal. So using that approach, you would get an error of up to 4% = 400Hz at 10kHz. Throw that one away. The short answer is that the only way you can do this is by making all counting, count stop and count read inside a CPLD, you can then transfer the result to a '51 for calculation and display.