Ensuring the confidentiality of data generated by IoT devices remains a pressing challenge due to their limited computational capabilities. In this work, we present an evolutionary approach to discover mathematical functions that enhance data masking through efficient cipher stream generation. The proposed method encodes data by applying a XOR operation between the input and a stream derived from these evolved functions, achieving lightweight encryption suited for resource-constrained environments. To maximize the masking effectiveness, the search process is guided by the optimization of the Shannon entropy of the resulting encoded strings. Experimental results show that the evolved functions can produce cryptograms with up to 2.6% higher entropy than those generated by the logistic map, and up to 1.2% compared to the Tinkerbell map. Moreover, the masked data successfully pass key statistical randomness tests from the NIST suite, including the Frequency, Cumulative Sums, and Block Frequency tests. These results highlight the potential of genetic programming design as a practical tool for lightweight cryptographic applications.