let us, for the sake of discussion, imagine a chip with a flssh sector/page size of 4k.

so, in the bottom 4k, you write something like

main(
...
func1()
...
func2()
...
func3()
...
func4()
...

then
in assembler (to avoid the dreaded C function pointers) and still in the bottom 4k
func1:
lcall 01000h
ret

func2:
lcall 02000h
ret

...

then make a separate compile (you may have to include the basics and strip them out) of REAL_func1 locating it at 0x1000

etc

then write a bootloader that can load individual pages/sectors.

Now, here is a problem. the OP states his code is 120k and the method descrived will, of course, leave gaps in memory.

Erik