Sounds like the solution is to write some math routines in assembler to cope with the extended range. One way would be to declare a type for your extended range. This type would amount to an array of 6 bytes (assuming 48bits). You would then pass pointers to these arrays to your assembler routines. Best hope that you're not doing anything more than +,-,*,/ !!!
Do a search on 32bit math routines and extend the code to cope with 48 bits.

Should be educational! Be sure to test thoroughly!