The Semantic Web (or Web of Data) represents the successful efforts towards linking and sharing data over the Web. The cornerstones of the Web of Data are RDF as data format and SPARQL as de-facto standard query language. Recent trends show the evolution of the Web of Data towards the Web of Things, integrating embedded devices and smart objects. Data stemming from such devices do not share a common format, making the integration and querying impossible. To overcome this problem, we present our approach to make embedded systems first-class citizens of the Web of Things. Our framework abstracts from individual deployments to represent them as common data sources in line with the ideas behind the Semantic Web. This includes the execution of arbitrary SPARQL queries over the data from a pool of embedded devices and/or external data sources. Handling verbose RDF data and executing SPARQL queries in an embedded network poses major challenges to minimize the involved processing and communication cost. We therefore present an in-network query processor aiming to push processing steps onto devices. We demonstrate the practical application and the potential benefits of our framework in a comprehensive evaluation using a real-world deployment and a range of SPARQL queries stemming from a common use case of the Web of Things.