Maintenance consumes a large part of the cost of software development which makes the optimization of that cost among the important issues in the world of software engineering. In this thesis we aim to optimize this cost by making these maintenances modular. To achieve this goal, we define transformations of program architectures that allow to transform a program to maintain into an architecture that facilitates the maintenance tasks required. We focus on transformation between architectures having dual modularity properties such as Composite and Visitor designpatterns. In this context, we define an automatic and reversible transformation based on refactoring between a program structured according to the Composite structure and its corresponding Visitor structure. This transformation is validated by generating a precondition which guarantees statically its success. It is also adapted to take into account the transformation of four variations of Composite pattern and it is then applied to JHotDraw program in which these four variations occur. We define also a reversible transformation in the Singleton pattern to benefit from optimization by introducing this pattern and flexibility by its suppression according to the requirements of the software user.
