The biomechanical materials are among the most complex mechanical systems. Most often, their microstructure are heterogeneous and random. For such systems, a gradient of porosity can be observed in the thickness direction but, in this case, none of the usual theories of porous materials can be applied. We propose a prior stochastic model of the tensor-valued elasticity field. There is a spatial gradient for the mean elasticity tensor and there are spatial statistical fluctuations in the thickness direction. The stochastic model which is constructed shows that the observed responses are effectively sensitive with respect to the values of the gradient and to the level of statistical fluctuations. In this sense, this stochastic model will be well adapted to perform its identification solving a stochastic inverse problem that is not the purpose of present paper.