The principle of "layered decoding" is extended to those codes not especially conceived for this practice, as to benefit of the increased convergence speed. Two different strategies are considered to solve the problem and the related architectures presented: one more straightforward, and based on the use for the soft output of the last value originated in a layer; the other based on the computation of the variation (or delta) of the soft output metrics to allow concurrent updates. Then, as in architecture-first approach, the performance are assessed for several widths of the layer, which remarks the robustness of the delta-mechanism to high parallelisation factors. As in the exact layered decoding, the average boost of two times in the convergence speed is shown.