The increasing adoption of the cloud computing paradigm has motivated a re definition of traditional software development methods. In particular, data storage management has received a great deal of attention, due to a growing interest in the challenges and opportunities associated to the NoSQL movement. However, appropriate selection, administration and use of cloud storage implementations remain a highly technical endeavor, due to large differences in the way data is represented, stored and accessed by these systems. This position paper motivates the use of model-driven techniques to avoid dependencies between high-level data models and cloud storage implementations. In this way, developers depend only on high-level data models, and then rely on transformation procedures to deal with particular cloud storage details, such as different APIs and deployment providers, and are able to target multiple cloud storage environments, without modifying their core data models.