Recently, traffic analysis and measurements have been used to characterize, from a security point of view, applications' and network behavior to avoid intrusion attempts, malware injections and data theft. Since most of the generated data traffic is from the embedded mobile devices, the analysis techniques have to cope on the one hand with the scarce computing capabilities and battery limitation of the devices, and on the other hand with tight performance constraints due to the huge generated traffic. In recent years, several machine learning approaches have been proposed in the literature, providing different levels of accuracy and requiring high computation resources to extract the analytic model from available training set. In this paper, we discuss a traffic analysis architecture that exploits FPGA technology to efficiently implement a hardware traffic analyzer on mobile devices, and a cloud infrastructure for the dynamic generation and updating of the data model based on ongoing mis-classification events. Finally, we provide a case study based on the implementation of the proposed traffic analyzer on a Xilinx Zynq 7000 architecture and Android OS, and show an overview of the proposed cloud infrastructure.