Application of an M-out-of-N redundancy architecture is a well-known measure for improving the reliability of safety systems. Most scientific papers addressing the reliability assessment of such systems consider a conventional homogeneous M-out-of-N redundancy architecture that is performed for identical channels. However, often in practice, an M-out-of-N redundancy architecture does not have identical channels. Reliability assessment of such heterogeneous systems (electrical/electronic and mechanical) with nonidentical channels and a combination of constant and nonconstant failure rates is considered in this paper. Such type of M-out-of-N redundancy architecture is introduced in this research as "asymmetrical redundancy". It can be used for enhancing the reliability of old mechanical systems or for reducing mutual influence of channels and increase of diagnostic coverage. This paper also presents a new "window-based Markov method" for PFDavg (average probability of failure on demand) and PFH (average frequency of dangerous failures) calculation for systems with an asymmetrical redundancy architecture and compares the results with those obtained by using the steady-state semi-Markov method and Monte-Carlo simulation. The applicability of the developed method is demonstrated in a simple case study.