In ubiquitous environments, context-aware applications need to monitor their execution context. They use middleware services such as context managers for this purpose. The space of monitorable entities is huge and each context-aware application has specific monitoring requirements which can change at runtime as a result of new opportunities or constraints due to context variations. The issues dealt with in this paper are 1) to guide context-aware application designers in the specification of the monitoring of distributed context sources, and 2) to allow the adaptation of context management capabilities by dynamically taking into account new context data collectors not foreseen during the development process. The solution we present, CA3M, follows the model-driven engineering approach for answering the previous questions: 1) designers specialised into context management specify context-awareness concerns into models that conform to a context-awareness meta-model, and 2) these context-awareness models are present at runtime and may be updated to cater with new application requirements. This paper presents the whole chain from the context-awareness model definition to the dynamic instantiation of context data collectors following modifications of context-awareness models at runtime