A small-signal model is the linear equivalent circuit that replaces a non-linear device for the purpose of signal analysis, valid for small swings about a fixed Operating point. It is the concrete output of Linearisation around an operating point: once you know the tangent slopes of the device’s characteristics at Q, you draw a little circuit of resistors and dependent sources that reproduces those slopes, and that circuit is the device as far as the signal is concerned.

Why it exists

A diode or transistor is non-linear, so it cannot be analysed with the resistor-and-source machinery of linear circuit theory. But after Linearisation around an operating point, the device’s response to a small signal is fully described by a handful of constants — derivatives of its laws evaluated at the bias point. A small-signal model packages those constants into an equivalent circuit so the rest of the analysis is pure linear-circuits work. This is Step 2 of Small-signal analysis: kill the DC sources, drop in the small-signal model, solve the linear circuit.

The technique is universal; only the parameters change

The procedure is identical across devices. What differs is the internal make-up of the equivalent circuit and the names of its parameters:

Every one of these parameters is a slope of a non-linear curve at Q; every one depends on the DC bias. That is why DC analysis (finding the Operating point) must come first: you cannot draw the small-signal model until you know the bias currents that set its parameter values. Master the diode case and the MOSFET and BJT cases are the same idea with more parameters.