In TF 2.1 (at least), you can create a model by passing to the outputs parameter a dictionary.

dictionary_as_model_output.py

def example1():
    # TensorFlow 2.1
    import tensorflow as tf
    import numpy as np

    Input_1 = tf.keras.layers.Input(shape=(1,))

    x = tf.keras.layers.Dense(100, activation='relu')(Input_1)

    targets = ["out1", "out2", "out3"]

    outputs = {}
    for target in targets:
        out1 = tf.keras.layers.Dense(1, activation='linear', name=target)(x)
        outputs[target] = out1

    model = tf.keras.models.Model(inputs=Input_1, outputs=outputs)
    model.compile(optimizer="rmsprop", loss={target: "mse" for target in targets},
                  metrics={target: ["mse", "accuracy"] for target in targets})

    data = np.array([[1], [2], [3], [4], [5]])

    predictions = model.predict(data)


if __name__ == '__main__':
    example1()