The Mobile Internet scenario encourages the design and development of context-aware applications that provide results depending on context information, such as the relative position of users, user preferences, device capabilities and available resources. A key requirement for the provisioning of context-aware applications is to give computer systems the ability to understand context information. Semantic languages are well suited to leverage the possibility to express, process and reason about context information and to facilitate knowledge sharing and interoperability among previously unknown entities accessing services from heterogeneous devices. However, the exploitation of semantic languages for the design and deployment of context-aware applications raises new challenges, mainly due to the high degree of heterogeneity that mobile devices exhibit in terms of computing power, memory, operating system, and supported software. Semantic languages require complex and heavyweight support facilities, e.g. metadata interpreters, reasoning engines and ontology repositories, that may not fit the capabilities of all user access devices, especially of the resource-limited ones. Novel solutions are required that are capable of transparently and dynamically adapting semantic support functionalities to the properties of the different user access devices. The paper proposes a novel middleware solution that exploits the visibility of two kinds of metadata, user/device profiles and policies, to tailor semantic support functionalities and that offers a wide set of mechanisms for providing on demand appropriate semantic support to mobile portable devices.