Stress is a general adaptation syndrome, caused by a person's perception of danger about an event. While the perception of an event as stressful varies for each individual, several studies show that public speaking is a source of anxiety and panic for most people. Moreover, the inability to prevent or manage these conditions affects their overall performance negatively. The literature depicts different systems developed in order to recognize the subject's stress level by monitoring physiological signals. However, they are specifically designed for laboratory environments and clinical analysis and are often characterized by several limitations. Therefore, aiming at developing a dailylife application able to help people prevent and handle anxiety conditions by improving their self-perception, we propose an embedded system for real-time biometric feature analysis. This system is able to handle all the steps from the acquisition to the stress and anxiety status classification. The whole framework is implemented on a Xilinx PYNQ-Z1 board, leveraging both the ARM processor and the FPGA. Taking advantage of the programmable logic brings to a system able to process data with higher performance and energy efficiency, helping to face the embedded application constraints.