We consider Secure Function Evaluation (SFE) in the client-server setting where the server issues a secure token to the client. The token is not trusted by the client and is \em not a trusted third party. We show how to take advantage of the token to drastically reduce the communication complexity of SFE and computation load of the server. Our main contribution is the detailed consideration of design decisions, optimizations, and trade-offs, associated with the setting and its strict hardware requirements for practical deployment. In particular, we model the token as a computationally weak device with small constant-size memory and limit communication between client and server. We consider semi-honest, covert, and malicious adversaries. We show the feasibility of our protocols based on a FPGA implementation.