This paper deals with soft-information based information reconciliation and data sifting for Quantum Key Distribution (QKD).We propose a novel composite channel model for QKD, which includes both a hard output quantum channel and a soft output classic channel. The metrics derived from the two channels are jointly processed at the receiver, exploiting capacity achieving soft-metric based iteratively decoded block codes. The performance of the proposed mixed-softmetric algorithms are studied via simulations as a function of the system parameters. The core ideas of the paper are: a) employing FEC coding as opposed to two-way communication for information reconciliation, b) exploiting all the available information for data processing at the receiver including information available from the quantum channel, since optimized use of this information can lead to significant performance improvement.