Model-based development aims at reducing the complexity of software development by the pervasive use of adequate models throughout the whole development process starting from early phases up to implementation. In this paper we present a conceptual framework to holistically classify developed models along different levels of abstraction. We do this by defining adequate abstractions for different development stages while ignoring the information that is not relevant at a particular development step or for a certain stakeholder. The abstraction is achieved in terms of the granularity level of the system under study (e. g. system, sub-system, sub-sub-system) and in terms of the information that the models contain (e. g. specification of functionality, description of architecture, deployment on specific hardware). We also present the relation between models that describe different perspectives of the system or are at different granularity levels. However, we do not address the process to be followed for building these models.