In a world populated by smart devices, the ability to handle large amounts of data becomes crucial. Big Data technologies are the most effective way to deal with such ungovernable, vast, and heterogeneous flows of information. Prompted by the rise of the Cloud Computing paradigm and the virtually infinite resources of data centers, Big Data solutions provide a scalable and efficient computation at a relatively low cost. However, due to the advent of IoT the amount of data generated and accumulated over the edge of the network is becoming alarming: the network infrastructure is not able to keep up with the data generation rates, becoming a de facto bottleneck. As a consequence, new trends are promoting the outsourcing of the computation from the cloud to near the data sources. In particular, Edge Computing is a computational paradigm focused on moving the computation on the devices at the extreme edges of the network, e.g., routers, sensors, or mobile phones. Notably, the task of performing a cohesive computation in such an unbounded, uneven, and unsteady environment is inherently complex. In order to solve this problem, we propose a paradigm shift in the design of data intensive applications: instead of trying to adapt the physical resources to the desired application, it is the application that has to adapt to the resources available. Given the high degree of heterogeneity, we also propose a declarative approach for building such complex deployment. In this context, the key idea is to abstract the device behaviour as a query, or a set of queries. Then, given a complex query that we want to execute over the edge framework, we (i) decompose the complex query into a cohesive query network, subsequently (ii) mapping the minimal resulting queries to the one exposed by the framework nodes.