To get more results or greater accuracy, computational scientists execute parallel or distributed applications, and try to scale these applications up. Clusters, Grids and Clouds are by nature different computing platforms: some span across multiple sites and multiple administrative domains, whereas others are single site/domain. As a consequence, scientists have to manage technical details related to both hardware and software constraints of each platform in order to execute their applications. From our point of view, scientists should focus on their research rather than dealing with platform considerations. In this article, we advocate for a system management framework that aims to automatically adapt both hardware and software resources to the applications' needs through a unique method. For each application, scientists describe the requirements through the definition of a Virtual Platform (VP) and a Virtual System Environment (VSE). Relying on these definitions, the framework is in charge of (i) the deployment and the configuration of the virtual platform upon the physical infrastructure, and (ii) the customization of the execution environment upon the former VP. Furthermore, we present a refinement of the concept of emulation and virtualization introduced by Goldberg. This allows scientists to define complex VPs relying on different concepts based on system virtualization (identity, partitioning, aggregation) and emulation (simple, abstraction). Leveraging several system management tools (such as OSCAR, the Grid'5000 toolkit, or XtreemOS), our current prototype follows a modular approach. We anticipate to be able to easily interface our architecture with recent cloud management solutions (such as OpenNebula or Eucalyptus).