Let's also take a look at the ScikitLearn.jl package, which defines a consistent API for fitting machine learning models and doing prediction. 

The following is how the FitBit type is defined:

""" `FitBit(model)` will behave just like `model`, but also supports
`isfit(fb)`, which returns true IFF `fit!(model, ...)` has been called """
mutable struct FitBit
    model
    isfit::Bool
    FitBit(model) = new(model, false)
end

function fit!(fb::FitBit, args...; kwargs...)
    fit!(fb.model, args...; kwargs...)
    fb.isfit = true
    fb
end

isfit(fb::FitBit) = fb.isfit

Here, we can see that the FitBit object contains a model object and that it adds a new functionality that keeps track of whether a model has been fitted or not:

@forward FitBit.model transform, predict, predict_proba, predict_dist, get_classes

It uses the @forward macro to delegate all the major functions, that is, transform, predict, and so on.