Random Linear Network Coding (RLNC) provides a theoretically efficient method for coding. The drawbacks associated with it are the complexity of the decoding and the overhead resulting from the coding vector. This adds to the overall energy consumption and is problematic for computational limited and battery driven platforms. In this work we present an approach to RLNC where the code is sparse and non-uniform. The sparsity allow for fast encoding and decoding, and the non- uniform protection of symbols enables recoding where the produced symbols are indistinguishable from those encoded at the source. The results show that the approach presented here provides a better trade- off between coding throughput and code overhead. In particular it can provide a coding overhead identical to RLNC but at significantly reduced computational complexity. It also allow for easy adjustment of this trade-off, which make it suitable for a broad range of platforms and applications. Finally it is easy to perform recoding and coding vectors can be efficiently represented.