Quasi-interpretations have shown their interest to deal with resource analysis of first order functional programs. There are at least two reasons to study the question of modularity of quasi-interpretations. Firstly, modularity allows to decrease the complexity of the quasi-inter\-pretation search algorithms. Secondly, modularity allows to increase the intentionality of the quasi-interpretation method, that is the number of captured programs. In particular, we take advantage of modularity conditions to extend smoothly quasi-interpretations to higher order programs. In this paper, we study the modularity of quasi-interpretations through the notions of constructor-sharing and hierarchical unions. We show that in the case of constructor-sharing and hierarchical unions, the existence of quasi-interpretations is no longer a modular property. However, we can still certify the complexity of programs.