IIRC, libraries like numpy and pytorch can already do that as they store the matrices as 1D arrays with information on things like the stride length (advancing to the next row). That allows you to implement operations like transposition by editing the stride length and other parameters without manipulating the content of the matrix array.