Large volumes of data from various sources are generated continuously in cities. The processing and analysis of these data play a key role in the implementation of initiatives for smart cities. In order to process urban Big Data, it is essential to use high-performance tools to accelerate processing and provide quick answers. However, this use is not trivial because Big Data tools are not interoperable and require from their users knowledge of parallel and distributed computing and databases. In this work, we compare popular open-source Big Data processing frameworks and propose a software system to abstract and facilitate their use in smart city applications. The architecture of the system is composed by an interface to specify dataflow models as well as services to interpret these models and instantiate them in different Big Data tools. An implementation of the system on top of a smart city platform is also addressed.