A novel strategy to encrypt covert information (code) via unitary projections into the null spaces of ill-conditioned eigenstructures of multiple host statistical distributions, inferred from incomplete constraints, is presented. The host pdf's are inferred using the maximum entropy principle. The projection of the covert information is dependent upon the pdf's of the host statistical distributions. The security of the encryption/decryption strategy is based on the extreme instability of the encoding process. A self-consistent procedure to derive keys for both symmetric and asymmetric cryptography is presented. The advantages of using a multiple pdf model to achieve encryption of covert information are briefly highlighted. Numerical simulations exemplify the efficacy of the model.