Much of Convolutional neural networks (CNNs)'s profound success lies in translation invariance. The other part lies in the almost infinite ways of arranging the layers of the neural network to make decisions in particular in computer vision problems, taking into account the whole image. This work proposes an alternative way to extend the pooling function, we named rank-order pooling, capable of extracting texture descriptors from images. Efforts to improve pooling layers or replace-add their functionality to other CNN layers is still an active area of research despite already a quite long history of architecture. Rank-order clustering is non-parametric, independent of geometric layout or image region sizes, and can therefore better tolerate rotations. Many related metrics are available for rank aggregation. In this article we present the properties of some of these metrics, their concordance indices and how they contribute to the efficiency of this new pooling operator.