UML is a visual modeling language for specifying, constructing, and documenting software intensive systems. UML diagrams play a central role in the whole software engineering process, starting from early analysis, through implementa- tion, to maintenance. One of the main weaknesses of UML, pointed out by surveys of UML use in industry, is the difficulty to express the context of a particular diagram or to enrich it to enhance its communication value. In fact, UML does not specify effective ways to express contextual information (e.g., project management indicators, quality indicators). To express this kind of information in a particular visual language, the Cognitive Dimensions framework proposes to use its secondary notation: the visual variables that are not used by the language. Unfortunately, in UML, this secondary notation is not controlled. The first contribution of this thesis showed - via an empirical study - that this results in a big variety of usages of these visual variables and mostly, in not effective ways. It also showed that UML practitioners acknowledge the importance of the secondary notation in their practice while confirming their lack of knowledge about the associated effective practices. In this thesis, we refer to the Semiology of Graphics (SoG) in order to target an optimal use of the UML secondary notation. The framework SoG-UML is the result of our adaptation of the SoG to UML via 61 guidelines about the use of the visual variables in this language. We have realized a first experiment to validate a part of these guidelines. Finally, we propose Fliplayers and an interactive keys, two Papyrus plug-ins that respectively implement the layers mechanism and the construction of "effective" keys.