"intercept/delay some of it"

This means data is received once by the bridge device and the device determines its response depending on the contents of the data. If the data transfer protocol itself has handshake, it's no problem. To delay the response, postpone to transfer acknowledge token.

However, if it depends on USB protocol level, it gets difficult.
On USB protocol, NAK is used to delay response.
IN-transaction (from device to host) is OK because the data comes from the other side.
Though OUT-transaction (from host to device) can also be delayed by NAK, but I don't know any device controller that can return NAK even if received OUT data is available from firmware. You may have to make an original USB engine on FPGA.

Tsuneo