@micione90 A dynamic module isn't a "true peak soft clipping" system. Since it has an attack time, it can't respect a cold hard true peak value.
But if you use a linear algorithm instead (i.e. tanh or any linear distortion formula of your choice) and upsample as @griffinboy said, you should be good.
As a general rule, a max of x8 upsampling is more than enough to respect a true peak standard.