Autonomous vehicles or devices like drones and missiles must make decisions without human assistance. Embedded inference engines, based for example on decision trees, are used to reach this goal which requires the respect of hardware and real-time constraints. This paper proposes a generic framework which can generate automatically the adequate hardware solution taking into account the real-time and the hardware constraints. Moreover, the proposed solution supports the case of faulty sensors which generate missing data. Experimental results obtained on FPGA and CPU suggest that it is better to consider missing data from learning phase instead of considering it just in the classification phase. Besides, the use of the pipeline technique provides a better latency, which is adequate for real-time applications.