Batched sparse (BATS) coding is a class of sparse random linear network coding scheme that achieves near-optimal tradeoff between temporal coding length and network throughput for file delivery over erasure networks. Existing BATS codes are mostly designed for equal error protection only. In practice, there are many applications that favor unequal error protection (UEP) for classifying and protecting the packets with different priorities. In this paper, we propose two classes of BATS code with UEP property, named weighted and expanding window BATS code. The asymptotic decoding probabilities of various types of input packets with the proposed coding schemes are analyzed via And-Or tree evaluation. Furthermore, the degree distributions for UEP BATS codes are optimized with linear programming and their performance is verified by computer simulations.