The number of Smart Cities in the world is growing very fast, and there is no indication that there will be a decrease in this growth soon, because a Smart City helps its managers and inhabitants to enjoy its resources and manage several of its aspects. City administrators are installing sensors and actuators in different parts of the city to collect data and react in real time to expected changes. It is up to computer programmers to design computer systems capable of processing all this data and making it available to inhabitants and administrators as information in an organized manner. To allow programmers to investigate the behavior of applications before they are deployed in the city, we have developed a three-layer testbed that helps programmers analyze performance in a controlled environment that can also be used for implementation. The testbed allows developers to distribute processing, including complex event/data streams, in the cloud, fog or edge. Although the testbed architecture is platform independent, we have implemented a reference version using free software. All the components used in the reference version were evaluated individually by their developers, but to verify the scalability of the integration, we developed several applications to evaluate the behavior of the architecture. Four applications for collecting and processing IoT data have been developed to illustrate how the testbed can guide programmers in choosing the best way to implement their applications. Based on the behavior of the developed applications, we created a taxonomy to classify applications for smart cities according to their characteristics and distribution possibilities to help the developer to implement their application.