The concept of Enterprise Application Integration (EAI) is widely used for integrating heterogeneous applications and systems via message-based communication. Typically, EAI servers provide a huge set of specific inbound and outbound adapters used for interacting with the external systems and for converting proprietary message formats. However, one major shortcoming in currently available products is the monolithic design of those adapters, resulting in performance deficits caused by the need for data independence. Further, also functional restrictions must be noticed here. In this paper, we give (1) a detailed problem characterization, followed by (2) a discussion of alternative data representations and adapter architectures, and (3) we introduce our model-driven DIEFOS (data independence, efficiency and functional flexibility using feature-oriented software engineering) approach. Finally, we analyze open research challenges.