Human Centered Computing is a novel paradigm to process context information of human being's environment in which computers are invisible to the subject, providing tools and services depending on the context of each individual. An increasing interest is growing regarding embedded computers since they offer advantages related to portability, dedicated tasks, invisibility, amongst others. However, a plenty of hardware platforms are in the market, so it is complicated to determine which the best for a particular need. Benchmarks make those tasks easier by performing measurements in hardware by running applications and performing comparisons. Nevertheless, they are thought to meet some quite particular kinds of applications. Moreover, if some benchmark applications have to be merged, i.e. voice or images processing, there are not schemes to correctly measure hardware platforms. On the other hand, requirements such as reliability and availability are not commonly assessed as a dependant set in hardware platforms. In this work, we propose a novel method to select hardware architectures for Human Centered Computing based on benchmarking, genetic algorithms, weighted sums and statistical distances in order to consider non-functional requirements.