Individually for each fraction.

 What do you mean :"should it be calculated for each peak individually?"

Do you have several peaks on your chromatogram?

SEC is a technique that enable determination of Mn, Mw, Mz by measuring the entire distribution of polymeric chains in the sample.

If you have two separate peaks then you have to do it separately.

But if you have bimodal signal then the situation is more comlplicated.

Actually, it's multimodal distribution and I'm using software for deconvolution, so an individual peaks are generated.

Your question seems simple, but the answer is complicated because the analysis problem is actually sophisticated.

A sample containing polymeric materials can be a mixture of homologous series (homopolymer), or of materials with varying architecture (branching), or varying composition (copolymer composition or a simple mixture), or all of the above.

There are several questions which could be considered: Are the two peaks the same polymer with the same composition and architecture ?

If the answer is yes, then the overall moments of the entire envelope represent the sample and its properties best. If, on the other hand, the goal is to understand the composition or polymerization effects, then having the two separate moments of the distributions can be very informative. 

If the answer is questionable, then analysis of the fractions needs to be done before an appropriate calculation can be made. The sample could be a mixture of different polymers, different branching, different colpolymer composition, or and combination of the above. Fractionation and analysis of the fractions (e.g. IR, UV, Viscometry, light scattering) will reveal the answer(s) to those questiions. This fractionation/analysis is sometimes done online multiple detection, or off line by separate analysis of fractions collected from the separation..Once the true nature of the sample is defined, then the appropriate calculations can be made.