Sorry for not being clear enough. I'll try to give more details.
Before doing the actual decimation, you filter the signal. The filter is a combinaison of digital multipliers, adders, and delay (a MAC in a DSP). First the 1-bit stream enter and say is multiplied by 2. You have then a 2-bits word. If after you integrate it (adder+dealy), you need more than 2 bits... and so on. Finally you can have the size you want. Note that if you undersize the words in the filter, you loose resolution. If you oversize, you have useless hardware.
Hope it helps
regards