For the last years, middleware industry, and particularly the CORBA project, have strongly integrated several aspects issued from databases: persistence, transaction management, security, etc. However, both most of object-oriented database management systems (OODBMS) and CORBA middleware have not integrated an essential component: the notion of views, which allows various users to see and manage data from different viewpoints. Views provide a semantic structuring of a schema of classes. A lot of application domains are interested in this feature; data mining is a well-known example. The paper focuses on CORBA and proposes complete software architecture in order to provide an efficient view mechanism for this middleware. This new software architecture provides solutions about evolution problems of available data. We experienced an original use of the IDL language and the IDL compiler, which consists of defining for each client (view) a subset of the global server IDL description.