In this paper we present a formal approach, based on meta-modelling and graph transformation, to the generation of environments for visual languages (VLs) supporting multiple views. The VL syntax is defined by means of a meta-model. Views for the VL are created by selecting the classes, associations, attributes and constraints that are part of them. Once the environment is generated, the view models are translated into a global, unique model where consistency checking and further analysis can be performed by means of graph grammars. These ideas have been newly implemented in the AToM3 tool.