optimizing CPU/FPU instructions

is it possible to code, say, an ADD that's faster than the actual instruction for ADD, only using other CPU functions (no physical access to the chip)? or are all CPU instructions always perfectly optimized?

e