Ok, i like the "many students must complicated..."

In this case:

Mix the psend and rd- througt an AND gate and the output of the gate to the memory.
Then you have a von newman structure (The program and data memory are the same)

You must put in the flash a code to download the code to put in the ram.
a) In the flash only the bootloader
b) In the flash the bootloader and the comon routines

If you need to use some interrupts then:
org int1
ljmp re_int1



org 1000h ;starts the ram
re_int1: ljmp your_code ;Redirected interrupt vector.

You must separate in two archives, the internal code and the external code. Download the internal code and after this starting the new_bootloader download the external code.

"And thats all"

This thread was discussed before, not exactly in the same terms of your case.

Search for bootloader.


.Sorry but... Why don't use a 64K Flash derivate? Then you can use 18 pins more for IO.