I have read some papers but i didn't get proper answers. In some cases, my level of theory is not exactly written. I will be very gladfull for any recommendation
You're adressing a basic issue with DFT: it's not systematically improvable, all that can be done is create a new functional and see what happens.
Optimization bond lengths: B3LYP is a hybrid functional which has been fitted to match ionization potentials and bond lengths from a molecule test set. Therefore, the expectation is that the bond lengths will be a distribution around the correct value, scaling factors can only be made for structurally similar systems.
Frequencies: for these you usually observe shifts instead of scaling factors, but that would also be material-specific, so I wouldn't transfer a shift calculated for a fullerene to a peptide.